[StepenkoAnatoli]

Fixes selftest-cli > multiple valid groups still run every matching test file, which failed the full suite under the commit gate's load: its child ran past a fixed 60 s timeout. The fix is to the test only.

What changed

• fetchproof/test/selftest-cli.test.mjs:
  • The two checks that run more than one group now run ordering witness instead of warc similarity. Both read only the kit, and the child takes a few seconds.
  • The child's spawnSync timeout is the harness's TEST_TIMEOUT (FETCHPROOF_TEST_TIMEOUT, default 60 s) instead of a fixed 60000 ms.
  • A failed check names the child's error (for example ETIMEDOUT) before its output, instead of a bare null !== 0.

Why

• The child for warc similarity takes 31-34 s on an idle PC. Under the gate's load on 2026-10-21 (the hand-run gate for the browser /stalled fix), it ran past 60 s, was killed with status null, and failed the suite.
• The fixed 60 s also ignored FETCHPROOF_TEST_TIMEOUT, which the gate's operator raises on a slow machine.
• Set aside:
  • Only raising the fixed timeout: that keeps a 30 s child in every run and just moves the line.
  • action-pins as the second group (1.8 s with ordering): it reads ../.github/scripts, outside the kit.
• Test mechanics, so no ADR (AGENTS.md rule 3). No README count change (3 tests before and after) and no CHANGELOG entry.

What you verified

• Red first, on the unchanged test. A preload held the two-group child for 35 s without using CPU. Result: 1 passed, 2 failed (null !== 0) in 124.8 s.
• Green on the fix, with the same 35 s hold: 3 passed, 0 failed in 89.7 s.
• Alone: 3 passed, 0 failed in 18.2 s while another suite ran. It took 64.5 s before the fix.
• Forced timeout (FETCHPROOF_TEST_TIMEOUT=5000, child held 8 s): both checks fail and say ETIMEDOUT.
• git diff --check is clean.
• Full suite through the renamed kit's gate, run by hand on 763ea36 (node fetchproof/bin/gate.mjs --gate commit): allowed in 2647 s. The installed hook is still the pre-rename kit's, which skips the suite on fetchproof/ paths. So the gate was run by hand first, and the commit then went through that hook.
• Rebased onto main after handoff: the OK line says how old the snapshot is; a stale page end to end (discussion #413, ADR-0187) #434 merged (30b77c9). handoff: the OK line says how old the snapshot is; a stale page end to end (discussion #413, ADR-0187) #434 changes none of the files this test runs, and selftest.mjs selftest-cli on the rebased head: 3 passed, 0 failed in 6.3 s. CI's suite is the full run on this head.

What I got wrong and fixed

• I first picked action-pins for its speed before checking what it reads. It was replaced by witness before the commit.

🤖 Generated with Claude Code